An empirical evaluation of rotation invariance of LDP feature for fingerprint matching using neural networks
Ravinder Kumar, Madasu Hanmandlu, Pravin Chandra · International Journal of Computational Vision and Robotics · 2014
Fingerprint-based individual authentication has been the most trusted and tested biometric among the biometrics traits. In the past two decades, many methods have been developed for fingerprint matching, but still there is a huge scope of improvement. This paper presents the rotation invariant fingerprint matching method, which is based on local directional pattern (LDP) features computed directly from grey values of a fingerprint image. For matching the extracted LDP histogram features, we have used single hidden layer feedforward neural networks (SLFNN). Six training algorithms namely, resilient propagation (RP), scaled conjugate gradient (SCG), gradient decent with all four variants [GD, GDM, GDA, GDX (refer to Table 2 for details)] are used for evaluating the matching performance and convergence time. The results show that the proposed features are invariant to the rotation and also suitable for fingerprint matching using SLFNN. The results also demonstrate that RP is better in performance than other investigated algorithms.